Re: [kernel-hardening] Re: [PATCH 4/4] refcount: Report failures through CHECK_DATA_CORRUPTION

[Kees Cook]

On Thu, Feb 9, 2017 at 2:27 AM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> On Wed, Feb 08, 2017 at 01:20:26PM -0800, Kees Cook wrote:
>
>> Ooooh, that is intense. And the trampolines (EX_REG_HANDLERs) are all
>> just there to catch whatever register gcc decides to stuff the value
>> into? *cover face* Sure, okay. :)
>
> Right, they shouldn't be big functions, but barring whole program LTO
> there's just no knowing which are unused.
>
>> I wonder how many existing WARN callsites could be repurposed to use this?
>
> At the very least all WARN/BUG instances with trivial @format argument
> that are inlined I think. For example, things like:
>
> static inline some_function()
> {
>         /* ... */
>         WARN(cond, "blah blah blah\n");
>         /* ... */
> }
>
> where the format has no arguments. Here we can out-of-line the printk()
> stuff, which, as is the purpose here, shrinks the size of the inline.

Unless there is some other unholy macros trick, I think we'd need a
separate "WARN_CONST" or something macro to do this (i.e.
WARN_CONST(const, const_str) instead of WARN(cond, fmt, ...)), since
detecting a single-item vararg in a macro is very weird/impossible to
do. Hrmm.

-Kees

-- 
Kees Cook
Pixel Security